Compare · read 21 July 2026
Both cost bases, published side by side.
Every "I replaced a $3,000 stack with AI for $200" post shares a shape: one number you cannot check, next to another number you cannot check. This page is the same comparison with the receipts attached. Ten licences at the tiers their vendors publish come to $1,769.91 a month. Our own infrastructure for the same work costs $59 a month. Neither figure includes a single person, and that gap is the whole argument.
So the numbers are built to be checked rather than believed. The licence side links to each vendor's own pricing page with the date it was read. Our side is broken into lines with a note on which ones are a real invoice and which are our own allocation. The first-party study underneath is published on gregshevchenko.com.
Three numbers on this page
$1,769.91 of licences at published tiers · $59 of our own infrastructure · eight categories of tool this stack does not replace.
Our side
$59 a month, and what that does not mean.
Agencies do not usually publish this. We are publishing it because the alternative is asking you to take the comparison on faith, and because the number is less flattering than it first looks: most of it is not even a bill we receive.
| Cost line | Per month | How solid the figure is |
|---|---|---|
| Mini-server electricity — it runs the content API, the model proxy and a local model | $8 | rough kWh estimate |
| Dedicated server — detection models, fact-check inference | $10 | our share of a host three products use |
| Flat-rate model subscription, content share | $5 | our share of one flat-rate seat |
| Paid API on the premium chain, occasional | $3 | typical month from usage logs |
| Scanner host | $8 | the invoice, in full |
| Fetcher host | $3 | our share of a shared host |
| The same dedicated server — scanner workloads | $10 | the same shared host, counted again for this product |
| Dashboard host | $5 | our share of a shared host |
| Regional search API | $5 | metered, about 5,000 queries |
| Regional generative API | $2 | free tier plus occasional paid |
| Model gateway — free provider pool | $0 | free tier |
| Fetcher upstreams — free tiers | $0 | free tier |
| Total | $59 | read 21 July 2026 |
The remaining $41 is our own allocation: shares of hosts that also run other work, a slice of one flat-rate subscription, and an electricity line that is an estimate from kilowatt-hours rather than a meter reading. Split it differently and you get a different number. We are telling you which part would move.
What it leaves out.
- People. The writer, the analyst, the operator who decides what to fix this month.
- The hardware itself beyond the shares listed — the hosts exist for other work too.
- Everything that is not measurement or production: sales, support, accounting.
The other side
$1,769.91 at published tiers.
10 licences, each read from the vendor's own page on 21 July 2026, each with the arithmetic that produced its monthly figure — a per-seat price multiplied by the seats, an annual licence divided by twelve. The full line-by-line table lives on our pricing page, and the measurement layer is compared vendor by vendor in the AI-visibility comparison.
An agency-tier AI-visibility seat and enterprise fact-check tooling are both real costs in a real stack, and neither vendor publishes a price. Our internal estimates put them near $2,494 a month together, which would nearly triple the licence side. We are not printing an estimate of somebody else's price: on 21 July our own research pass found five errors in our competitor sheet, four of them in our favour, and this is the error that is fastest for a reader to disprove.
The part neither number covers
The distance between $59 and a working programme is a person.
Put the two figures next to each other and the obvious reading is that we mark up infrastructure by a large multiple. The honest answer is that infrastructure is not what anyone is buying. A vendor charging you $399 a month also serves your account for a fraction of it; the difference here is that we are showing our fraction.
What the money actually buys is the loop between the two numbers. Something has to read the measurement and decide which of forty gaps is worth this month. Something has to draft the page, catch the fabricated statistic before it publishes, check the claim against a source, and put the result back in front of the engines to see whether the citation moved. Every one of those steps has a gate, and every gate was built because it failed once. That is the expensive half, it does not appear on any invoice, and it is the reason a stack of licences and a working programme are different things.
Still on your card
What this stack does not replace.
Taken from the open gaps in our own catalogue. A roll-up that consolidates everything is a roll-up that has not been used in anger.
| If you need | You keep paying for |
|---|---|
| A writer-facing WYSIWYG editor | Google Docs, Notion, or your CMS editor |
| Daily rank tracking in classical search | Ahrefs, Semrush, or a dedicated rank tracker |
| Backlink analysis and outreach | Ahrefs or Semrush |
| Session heatmaps and replays | Hotjar, Clarity, or similar |
| Post-publish web analytics | GA4, Plausible, or your existing analytics |
| Slack alerts when visibility drops | a manual check of the dashboard |
| Sentiment analysis on AI mentions | nobody, in our stack — read the answers yourself |
| PR and journalist mention tracking | a media-monitoring tool |
The consolidation is real but narrow: it is the measurement-and-production loop, not the whole toolbox. If your problem is backlinks or conversion-rate optimisation, this is not the page that solves it.
Method
How to check this yourself.
The licence half is fully reproducible and takes about twenty minutes. Open each vendor page linked from the pricing table, find the tier named in the row, and compare. Where a row multiplies — a per-seat price by ten seats, an annual licence by one twelfth — the arithmetic is printed next to it, so a disagreement about team size is visible rather than buried in a total.
Two details will trip a naive check. First, some vendors render the price in the browser rather than in the page source, so a script that reads the raw HTML sees nothing; those rows say renders in the browser. Second, an annual-billing price and a monthly price for the same tier can differ by a fifth, and vendors do not always show both at once. Every row here says which basis it used.
Our half you cannot reproduce, and pretending otherwise would be dishonest: nobody outside can see our hosting invoices. What you can do is check it for internal consistency. The lines that are shares of shared hosts are marked, the estimate is marked as an estimate, and the two halves of the split add up to the total. If a number on this page ever stops adding up, that is a reportable error and we would rather hear about it than not.
Prices in this category move faster than comparison pages get updated, which is why every figure carries the day it was read rather than a year. If you are reading this months later, treat the dates as the expiry label they are, and re-check the two or three lines that matter most to your decision.
Licence side
The vendor’s own page
Ten published tiers, each read on 21 July 2026 and linked to the page it came from. Prices that render only in a browser are marked as such.
Our side
Our own cost notes
Twelve lines from our internal operating-cost notes, split into what is billed and what we allocate. Read 21 July 2026.
The gaps
Our own catalogue
The "does not replace" list is the open-gap register we maintain internally, filtered to what is still open across the whole stack.
Questions buyers actually ask
Cost questions, answered directly.
What does the $59 actually cover?
Servers, electricity and API calls for the measurement and production side, read 21 July 2026. Of that, $18 is a bill or a metered API and $41 is our own allocation of hosts shared with other work, including one line that is a rough electricity estimate. It covers no people at all.
So you charge $2,497 for something that costs you $59?
No. $59 is what the machines cost. The price is the operator who runs them, the loop that turns a measurement into a published page, and the accountability for the result. A licence you buy for $399 also costs its vendor a fraction of that to serve — the difference is that we are showing you our fraction.
Why is the licence side $1,769.91 and not the $3,864.91 you quote internally?
Because two of the bundle lines are prices no vendor publishes: an agency-tier AI-visibility seat and enterprise fact-check tooling. We hold internal estimates for both, worth about $2,494 a month together, and we will not print an estimate of somebody else's price on a public page. On 21 July our own research pass found five errors in our competitor sheet, four of them flattering to us.
Could I just build this myself for $59?
The infrastructure, yes — most of it is open source or a free tier. What takes the time is the part that is not in the bill: the evaluation harness that decides whether a draft is publishable, the gates that stop a wrong number from shipping, and the person who reads the measurement and picks what to fix. That took us two years and it is still the expensive half.
Does this replace my whole marketing stack?
No, and the list of what it does not replace is on this page. Rank tracking, backlinks, heatmaps, web analytics and media monitoring all stay where they are. What consolidates is the measurement-and-production loop, not the entire toolbox.
How often do these numbers change?
Vendor prices are re-read on a schedule and every figure on this page carries the date it was read — 21 July 2026 for the licence side, 21 July 2026 for ours. A comparison page without a date is a rumour, and this category re-prices faster than most pages get updated.
Want the version with your own numbers in it?
Thirty minutes that opens with a citation map for your category across the engines, and an honest read on whether any subscription here — including somebody else's — is the right purchase for you this quarter.